task IO accounting: provide distinct tgid/tid I/O statistics
[safe/jmp/linux-2.6] / fs / proc / base.c
index c5e412a..a891fe4 100644 (file)
@@ -56,6 +56,7 @@
 #include <linux/init.h>
 #include <linux/capability.h>
 #include <linux/file.h>
+#include <linux/fdtable.h>
 #include <linux/string.h>
 #include <linux/seq_file.h>
 #include <linux/namei.h>
@@ -126,6 +127,25 @@ struct pid_entry {
                NULL, &proc_single_file_operations,     \
                { .proc_show = &proc_##OTYPE } )
 
+/*
+ * Count the number of hardlinks for the pid_entry table, excluding the .
+ * and .. links.
+ */
+static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
+       unsigned int n)
+{
+       unsigned int i;
+       unsigned int count;
+
+       count = 0;
+       for (i = 0; i < n; ++i) {
+               if (S_ISDIR(entries[i].mode))
+                       ++count;
+       }
+
+       return count;
+}
+
 int maps_protect;
 EXPORT_SYMBOL(maps_protect);
 
@@ -195,12 +215,32 @@ static int proc_root_link(struct inode *inode, struct path *path)
        return result;
 }
 
-#define MAY_PTRACE(task) \
-       (task == current || \
-       (task->parent == current && \
-       (task->ptrace & PT_PTRACED) && \
-        (task_is_stopped_or_traced(task)) && \
-        security_ptrace(current,task) == 0))
+/*
+ * Return zero if current may access user memory in @task, -error if not.
+ */
+static int check_mem_permission(struct task_struct *task)
+{
+       /*
+        * A task can always look at itself, in case it chooses
+        * to use system calls instead of load instructions.
+        */
+       if (task == current)
+               return 0;
+
+       /*
+        * If current is actively ptrace'ing, and would also be
+        * permitted to freshly attach with ptrace now, permit it.
+        */
+       if (task->parent == current && (task->ptrace & PT_PTRACED) &&
+           task_is_stopped_or_traced(task) &&
+           ptrace_may_access(task, PTRACE_MODE_ATTACH))
+               return 0;
+
+       /*
+        * Noone else is allowed.
+        */
+       return -EPERM;
+}
 
 struct mm_struct *mm_for_maps(struct task_struct *task)
 {
@@ -211,7 +251,8 @@ struct mm_struct *mm_for_maps(struct task_struct *task)
        task_lock(task);
        if (task->mm != mm)
                goto out;
-       if (task->mm != current->mm && __ptrace_may_attach(task) < 0)
+       if (task->mm != current->mm &&
+           __ptrace_may_access(task, PTRACE_MODE_READ) < 0)
                goto out;
        task_unlock(task);
        return mm;
@@ -478,7 +519,7 @@ static int proc_fd_access_allowed(struct inode *inode)
         */
        task = get_proc_task(inode);
        if (task) {
-               allowed = ptrace_may_attach(task);
+               allowed = ptrace_may_access(task, PTRACE_MODE_READ);
                put_task_struct(task);
        }
        return allowed;
@@ -722,7 +763,7 @@ static ssize_t mem_read(struct file * file, char __user * buf,
        if (!task)
                goto out_no_task;
 
-       if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
+       if (check_mem_permission(task))
                goto out;
 
        ret = -ENOMEM;
@@ -748,7 +789,7 @@ static ssize_t mem_read(struct file * file, char __user * buf,
 
                this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
                retval = access_process_vm(task, src, page, this_len, 0);
-               if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) {
+               if (!retval || check_mem_permission(task)) {
                        if (!ret)
                                ret = -EIO;
                        break;
@@ -792,7 +833,7 @@ static ssize_t mem_write(struct file * file, const char __user *buf,
        if (!task)
                goto out_no_task;
 
-       if (!MAY_PTRACE(task) || !ptrace_may_attach(task))
+       if (check_mem_permission(task))
                goto out;
 
        copied = -ENOMEM;
@@ -864,7 +905,7 @@ static ssize_t environ_read(struct file *file, char __user *buf,
        if (!task)
                goto out_no_task;
 
-       if (!ptrace_may_attach(task))
+       if (!ptrace_may_access(task, PTRACE_MODE_READ))
                goto out;
 
        ret = -ENOMEM;
@@ -1181,6 +1222,81 @@ static const struct file_operations proc_pid_sched_operations = {
 
 #endif
 
+/*
+ * We added or removed a vma mapping the executable. The vmas are only mapped
+ * during exec and are not mapped with the mmap system call.
+ * Callers must hold down_write() on the mm's mmap_sem for these
+ */
+void added_exe_file_vma(struct mm_struct *mm)
+{
+       mm->num_exe_file_vmas++;
+}
+
+void removed_exe_file_vma(struct mm_struct *mm)
+{
+       mm->num_exe_file_vmas--;
+       if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
+               fput(mm->exe_file);
+               mm->exe_file = NULL;
+       }
+
+}
+
+void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
+{
+       if (new_exe_file)
+               get_file(new_exe_file);
+       if (mm->exe_file)
+               fput(mm->exe_file);
+       mm->exe_file = new_exe_file;
+       mm->num_exe_file_vmas = 0;
+}
+
+struct file *get_mm_exe_file(struct mm_struct *mm)
+{
+       struct file *exe_file;
+
+       /* We need mmap_sem to protect against races with removal of
+        * VM_EXECUTABLE vmas */
+       down_read(&mm->mmap_sem);
+       exe_file = mm->exe_file;
+       if (exe_file)
+               get_file(exe_file);
+       up_read(&mm->mmap_sem);
+       return exe_file;
+}
+
+void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
+{
+       /* It's safe to write the exe_file pointer without exe_file_lock because
+        * this is called during fork when the task is not yet in /proc */
+       newmm->exe_file = get_mm_exe_file(oldmm);
+}
+
+static int proc_exe_link(struct inode *inode, struct path *exe_path)
+{
+       struct task_struct *task;
+       struct mm_struct *mm;
+       struct file *exe_file;
+
+       task = get_proc_task(inode);
+       if (!task)
+               return -ENOENT;
+       mm = get_task_mm(task);
+       put_task_struct(task);
+       if (!mm)
+               return -ENOENT;
+       exe_file = get_mm_exe_file(mm);
+       mmput(mm);
+       if (exe_file) {
+               *exe_path = exe_file->f_path;
+               path_get(&exe_file->f_path);
+               fput(exe_file);
+               return 0;
+       } else
+               return -ENOENT;
+}
+
 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
        struct inode *inode = dentry->d_inode;
@@ -2260,29 +2376,82 @@ static int proc_base_fill_cache(struct file *filp, void *dirent,
 }
 
 #ifdef CONFIG_TASK_IO_ACCOUNTING
-static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
-{
+static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
+{
+       u64 rchar, wchar, syscr, syscw;
+       struct task_io_accounting ioac;
+
+       if (!whole) {
+               rchar = task->rchar;
+               wchar = task->wchar;
+               syscr = task->syscr;
+               syscw = task->syscw;
+               memcpy(&ioac, &task->ioac, sizeof(ioac));
+       } else {
+               unsigned long flags;
+               struct task_struct *t = task;
+               rchar = wchar = syscr = syscw = 0;
+               memset(&ioac, 0, sizeof(ioac));
+
+               rcu_read_lock();
+               do {
+                       rchar += t->rchar;
+                       wchar += t->wchar;
+                       syscr += t->syscr;
+                       syscw += t->syscw;
+
+                       ioac.read_bytes += t->ioac.read_bytes;
+                       ioac.write_bytes += t->ioac.write_bytes;
+                       ioac.cancelled_write_bytes +=
+                                       t->ioac.cancelled_write_bytes;
+                       t = next_thread(t);
+               } while (t != task);
+               rcu_read_unlock();
+
+               if (lock_task_sighand(task, &flags)) {
+                       struct signal_struct *sig = task->signal;
+
+                       rchar += sig->rchar;
+                       wchar += sig->wchar;
+                       syscr += sig->syscr;
+                       syscw += sig->syscw;
+
+                       ioac.read_bytes += sig->ioac.read_bytes;
+                       ioac.write_bytes += sig->ioac.write_bytes;
+                       ioac.cancelled_write_bytes +=
+                                       sig->ioac.cancelled_write_bytes;
+
+                       unlock_task_sighand(task, &flags);
+               }
+       }
+
        return sprintf(buffer,
-#ifdef CONFIG_TASK_XACCT
                        "rchar: %llu\n"
                        "wchar: %llu\n"
                        "syscr: %llu\n"
                        "syscw: %llu\n"
-#endif
                        "read_bytes: %llu\n"
                        "write_bytes: %llu\n"
                        "cancelled_write_bytes: %llu\n",
-#ifdef CONFIG_TASK_XACCT
-                       (unsigned long long)task->rchar,
-                       (unsigned long long)task->wchar,
-                       (unsigned long long)task->syscr,
-                       (unsigned long long)task->syscw,
-#endif
-                       (unsigned long long)task->ioac.read_bytes,
-                       (unsigned long long)task->ioac.write_bytes,
-                       (unsigned long long)task->ioac.cancelled_write_bytes);
+                       (unsigned long long)rchar,
+                       (unsigned long long)wchar,
+                       (unsigned long long)syscr,
+                       (unsigned long long)syscw,
+                       (unsigned long long)ioac.read_bytes,
+                       (unsigned long long)ioac.write_bytes,
+                       (unsigned long long)ioac.cancelled_write_bytes);
 }
-#endif
+
+static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
+{
+       return do_io_accounting(task, buffer, 0);
+}
+
+static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
+{
+       return do_io_accounting(task, buffer, 1);
+}
+#endif /* CONFIG_TASK_IO_ACCOUNTING */
 
 /*
  * Thread groups
@@ -2345,7 +2514,7 @@ static const struct pid_entry tgid_base_stuff[] = {
        REG("oom_adj",    S_IRUGO|S_IWUSR, oom_adjust),
 #ifdef CONFIG_AUDITSYSCALL
        REG("loginuid",   S_IWUSR|S_IRUGO, loginuid),
-       REG("sessionid",  S_IRUSR, sessionid),
+       REG("sessionid",  S_IRUGO, sessionid),
 #endif
 #ifdef CONFIG_FAULT_INJECTION
        REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
@@ -2354,7 +2523,7 @@ static const struct pid_entry tgid_base_stuff[] = {
        REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
 #endif
 #ifdef CONFIG_TASK_IO_ACCOUNTING
-       INF("io",       S_IRUGO, pid_io_accounting),
+       INF("io",       S_IRUGO, tgid_io_accounting),
 #endif
 };
 
@@ -2489,10 +2658,9 @@ static struct dentry *proc_pid_instantiate(struct inode *dir,
        inode->i_op = &proc_tgid_base_inode_operations;
        inode->i_fop = &proc_tgid_base_operations;
        inode->i_flags|=S_IMMUTABLE;
-       inode->i_nlink = 5;
-#ifdef CONFIG_SECURITY
-       inode->i_nlink += 1;
-#endif
+
+       inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
+               ARRAY_SIZE(tgid_base_stuff));
 
        dentry->d_op = &pid_dentry_operations;
 
@@ -2682,6 +2850,9 @@ static const struct pid_entry tid_base_stuff[] = {
 #ifdef CONFIG_FAULT_INJECTION
        REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
 #endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+       INF("io",       S_IRUGO, tid_io_accounting),
+#endif
 };
 
 static int proc_tid_base_readdir(struct file * filp,
@@ -2720,10 +2891,9 @@ static struct dentry *proc_task_instantiate(struct inode *dir,
        inode->i_op = &proc_tid_base_inode_operations;
        inode->i_fop = &proc_tid_base_operations;
        inode->i_flags|=S_IMMUTABLE;
-       inode->i_nlink = 4;
-#ifdef CONFIG_SECURITY
-       inode->i_nlink += 1;
-#endif
+
+       inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
+               ARRAY_SIZE(tid_base_stuff));
 
        dentry->d_op = &pid_dentry_operations;